Design And Evaluation Of A Click-through Rate Prediction Model Based On Deep Preference Network

Click-through Rate prediction is a core research issue in business scenarios such as recommendation systems and online advertisements,and plays a pivotal role in the entire Internet field.In recent years,artificial intelligence and deep learning have made many advances in the fields of computer vision and natural language processing.It makes Internet companies and relevant research institutes have made many explorations and studies on the application of deep learning to Click-through Rate prediction,and have achieved a series of excellent results.This paper summarizes the research results of a series of deep learning in recent years on the prediction of Click-through Rate,finds that many of the current Click-through Rate prediction models based on deep learning still have many traps and deficiencies in the mining of personalized preferences for users.There are still a lot of room for optimization and improvement,especially in the use of personalized historical behavior feature of users.Based on this background,this paper proposes a Click-through Rate prediction model based on deep preference network by studying various unique network structures of deep learning in computer vision and natural language processing,as well as its own experiments and explorations on various network structures.The main content of the paper can be summarized as the following three aspects:Firstly,this paper summarizes the various Click-through Rate prediction models based on deep learning techniques that have emerged in recent years,and classifies and summarizes each model according to its structural characteristics and application scope.On this basis,this paper elaborates on the related technology of DNN-based Click-through Rate prediction model and the theory of Attention-related technology.Secondly,this paper proposes a deep preference network for mining user personalized preference information,and builds a Click-through Rate prediction model.In this paper,the Self-Attention and Multi-Head Attention structures,which have many good applications in the field of computer vision and natural language processing,are introduced into the deep learning model of Click-through Rate prediction through experimental improvement.Based on this,this paper combines the advantages of traditional DNN structure as the main Click-through Rate prediction model,and proposes a Click-through Rate prediction model based on depth preference network.Thirdly,the design experiment proves that the proposed model has superiority in prediction accuracy,structural design rationality and model interpretability.The experimental results show that,in terms of accuracy of model prediction,the proposed DPN-based Clickthrough Rate prediction model's AUC achieves 0.8038,which is 0.4%-1.37% higher than other models.In terms of the model's ability to utilize the characteristics of the behavior sequence,the prediction effect of the proposed model can be improved with the increase of the length of the behavior sequence.In terms of the validity of the sub-structure of the model,the main substructure of the proposed model plays a positive role in promoting the overall prediction effect of the model.In terms of the interpretability of the Embedding vector generated by the model,the model proposed in this paper has better interpretability and applicability than the comparative model.